Determining Vegetation Indices From Solar and Photosynthetically Active Radiation Fluxes
description
Transcript of Determining Vegetation Indices From Solar and Photosynthetically Active Radiation Fluxes
-
n i
ct
*, T
/NOA
d form
abilit
near
data from 11 flux tower locations in 5 vegetation types (desert grassland, temperate grasslands, crops, deciduous forests, and pine
Agricultural and Forest Meteorologyforest) were collected across the United States. Vegetation indices were derived using solar radiation and photosynthetically active
radiation (PAR) measured above the vegetation canopy throughout the year. The normalized-difference vegetation index (NDVI)
estimated using the 30-min data was then used to quantify the LAI of the vegetation types at the various sites. The exponential
function between LAI and NDVI indicated a non-linear relationship with the maximum tower-derived NDVI/LAI about 0.82/4.5 for
corn, 0.85/6 for soybean, 0.6/20.8/4 for grasslands, and 0.81/7 for forest. Each vegetation type and environment exhibited unique
seasonal and annual signatures of NDVI/LAI. The NDVI/LAI from the flux towers compared well with the Moderate Resolution
Imaging Spectroradiometer (MODIS) data derived at 1-km resolution and derived LAI showed excellent agreement with
measurements in corn/soybean crops. These results encourage the use of real-time single point measurements of vegetation
spectral indices in characterizing vegetation for routine plant-environment models.
# 2007 Elsevier B.V. All rights reserved.
Keywords: Remote sensing; Leaf area; Vegetation indices; Surface energy fluxes
1. Introduction
Since 1997 the NOAA-ATDD-GEWEX program has
been conducting continuous detailed measurements of
surface fluxes of energy, water, and carbon as well as
other local vegetation-climate state variables using
collections of flux tower locations in various vegetation
types across the United States. This program seeks to
provide continuous field observations for understanding
and characterizing the influence of soil-vegetation
environments on the energy, water and carbon exchange
processes at the earths surface. These exchange
processes are affected, in large part, by variations of
vegetation properties such as the canopy height and leaf
area over the soil surface. Information about the
seasonal distribution of vegetation is important for
determining the contributions of vegetation to the
surface fluxes of energy, water, and carbon as the leaf
area increases over the soil during periods of vegetation
growth, as well as the dominance of the bare soil and
residue layers when vegetation is reduced by seasonal
senescence or a field management strategy.
Although many instruments exist for determining the
leaf area index (LAI) of vegetation, they are often
laborious and costly to run continuously at flux tower
locations. Unlike most flux sensors that are designed to
run in all weather conditions, most ground-based
* Corresponding author. Tel.: +1 865 576 1249;
fax: +1 865 576 1327.
E-mail address: [email protected] (T.B. Wilson).
0168-1923/$ see front matter # 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.agrformet.2007.04.001Determining vegetatio
photosynthetically a
T.B. Wilson
Atmospheric Turbulence and Diffusion Division
Received 21 June 2006; received in revise
Abstract
The objective of this study was to quantify the seasonal vari
for use in soilvegetationatmosphere exchange models usingndices from solar and
ive radiation fluxes
.P. Meyers
A, P.O. Box 2456, Oak Ridge, TN 37831, USA
6 February 2007; accepted 2 April 2007
y of vegetation spectral indices to deduce leaf area index (LAI)
-real-time and archived flux tower radiation data. The 30-min
www.elsevier.com/locate/agrformet
144 (2007) 160179
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 161instruments of LAI operate largely in non-precipitation
conditions. Measurements of LAI using traditional
sensors (e.g., LAI-2000; Li-Cor, 1990) also require
multiple in-field sampling, making them unsuitable to
operate at the usual single point where most sensors are
mounted to flux towers. Similarly, measuring LAI
manually is becoming impractical for regional mon-
itoring networks such as found in the GEWEX program
where field sites are expanding to distant locations in
several vegetation regions across the US. While remote
sensing satellites offer a good alternative for obtaining
the description of regional and global vegetation
(Lotsch et al., 2003; Chen et al., 2002; Hall et al.,
1996, 2003; Knyazikhin et al., 1998; Spanner et al.,
1990; White et al., 1997), data derived from satellites
such as the Moderate Resolution Imaging Spectrometer
(MODIS) are provided once every 816-day period
(Huete et al., 2002; Lotsch et al., 2003), and require
calibration to represent site-specific vegetation condi-
tions. In addition, satellite data can be expensive and
difficult to process and contain occasional spurious
inferences due to aerosols, cloud cover, and the surface
bidirectional reflectance distribution function (BRDF)
(Disney et al., 2004; Berk et al., 1998).
The high quality radiation flux data being compiled
from11AmerifluxGEWEX sites reveal that it is possible
to calculate spectral vegetation indices (VI) using the
regular radiation flux measurements immediately above
the vegetation. Acceptable VI estimates were obtained
from the reflectance in the visible (VIS, 400700 nm)
and near-infrared (NIR, 7003000 nm) spectral bands of
the landvegetation surface derived from tower mea-
surements of incoming and outgoing global solar and
photosynthetically active radiation (PAR) fluxes. In
particular, daily values of VI were obtained by
aggregating values estimated from the 30-min radiation
data during the daytime, clear sky conditions from 10:00
to 14:00 h. During this period the radiation fluxes
measured at a fixed point from the tower above the
canopy corresponded to solar zenith angles < 30, anassumption which has been sufficient to limit the effects
of BRDFonVIS andNIR (Disney et al., 2004). Values of
the VI were then used to obtain nominal LAI values
relevant for the various GEWEX flux tower sites. Our
tower-derivedLAIdata appear satisfactory for evaluating
satellite data relevant in characterizing small-scale
vegetation features at flux tower sites. Disney et al.
(2004) recently conducted similar pointmeasurements of
albedo over crops in theUK thatwere successfully scaled
up to the 1-km resolution of the Moderate Resolution
Imaging Spectrometer (MODIS). The point measure-
ments of albedo agreedwithin about 3%with theMODISvalues across the visible, shortwave IR and near IR parts
of the spectrum. In spite of such good comparison,
remote sensing satellites still suffer cloud contamination
and do not provide daily products such asVI and LAI in a
continuous time series to always depict vegetation
conditions accurately during the growing season.
Furthermore, the pixel size of MODIS products often
covers large areas with different vegetation types. In this
situation, more site-specific information is needed to
specify the proper spatial scales and pixel size in using
MODIS data to study small-scale ecosystems. Tower-
based measurements offer the potential to overcome
some of these challenges with satellite products by
providing vegetation information above the canopy that
combines contributions of the vegetation of interest and
the background (such as bare soil or plant residue) at field
levels. Thus, estimating site-specific LAI from such
tower measurements offers the possibility of providing
consistent vegetation characterization in modeling
energy, water, and carbon fluxes of soilvegetation
atmosphere systems.
Themain objectives of this study are to: (1) determine
the annual variability of vegetation indices at 11 locations
of flux towers relevant for evaluating the interactions of
the landvegetationatmosphere system; (2) provide
simple, practical estimates of the temporal changes in the
vegetation canopy cover using continuous, hourly
measurements of global solar and PAR fluxes at a fixed
point above thevegetation; (3) evaluate the tower-derived
vegetation indices against products of satellite remote
sensing and a selection of field measurements. The study
was done with the aim of providing consistent LAI
parameters required for landvegetation models; thus
LAI was calculated as a function of the normalized
difference vegetation index (NDVI), which is the most
widely used reflectance vegetation index (Pontailer et al.,
2003; Qi et al., 2000; Spanner et al., 1990). The
performance of the tower-derived NDVI and LAI based
on wide bandwidths of VIS (400700 nm) and NIR
(7003000 nm) were compared with the MODIS NDVI
and LAI products, which are derived using the narrow
bands of red (620670 nm) and NIR (841876 nm);
other studies have also evaluated these MODIS products
in describing the seasonal dynamics of various vegetation
types (Ferreira et al., 2003; Huete et al., 2002).
2. Methods
2.1. Estimation of spectral vegetation indices
From the radiation spectrum, let RVIS denote thereflectance of the visible radiation (400700 nm) and
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179162the RNIR denote the reflectance of the near-infraredradiation (7003000 nm), then
RVIS PARoutPARin
(1)
can be approximated by measuring PARin and PARout,
the respective incoming and outgoing PAR fluxes
(mmol m2 s1) above the vegetation canopy. In otherto calculate RNIR, the global solar (SOLR) radiation (inW m2) was partitioned into downward components ofvisible (VISin) and near-infrared (NIRin) wavebands as
follows (Weiss and Norman, 1985; Moon, 1940):
VISin 0:45 SOLR (2)and
NIRin 0:55 SOLR (3)Using this partition, the VIS reflected upward (VISout)
was calculated as
VISout RVIS VISin (4)Using the measurements of the reflected global solar
(SOLRout) radiation (in W m2), the NIR reflected
upward (NIRout) was calculated as
NIRout SOLRout VISout (5)From Eqs. (3) and (5), RNIR is calculated as
RNIR NIRoutNIRin
(6)
Using Eqs. (1) and (6), NDVI is calculated as
NDVI RNIR RVISRNIR RVIS (7)
Eq. (7) is based on the fact that plant canopies have
stronger absorption (8590%) and lower reflectance and
transmittance (510%) in the visible waveband; in
contrast, they have much greater reflectance and trans-
mission and lower absorption in the near infrared
radiation band (Campbell and Norman, 1998). Because
the NDVI is derived from the entire field of view, it may
not remove the ground surface that forms part of the
background to the vegetation canopy (Richardson and
Wiegand, 1977). Correction schemes have been devel-
oped to adjust the NDVI values from Eq. (7)(Carlson
et al., 1995; Campbell and Norman, 1998), which may
not reach zero for zero vegetation cover due to different
soil reflectance factors in the two wave bands or due
atmospheric effects; examples include the weighted
difference vegetation index (WDVI) and spectral mix-
ture analysis (SMA) (Clevers, 1989; Hall et al., 1995,
2003; Peddle et al., 1999, 2001; Peddle and Johnson,2000), and soil-adjusted vegetation index (SAVI) (Qi
et al., 1994).
However, in this study none of these correction
schemes was applied to the NDVI values from Eq. (7),
because of the lack of LAI measurements at the various
flux tower locations. To estimate LAI from Eq. (7), a
negative exponential function in line with the adjusted
NDVI and canopy fraction equations reported by
Campbell and Norman (1998) (see Eqs. (15.32) and
(15.33)) was used to relate LAI (in m2 m2) to NDVI:
LAI K log
NDVImax NDVINDVImax NDVImin
(8)
where NDVImin NDVI
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 163
itude
in-s
049000240000100
80240001200
6000000120002400
0036000260002400Oak Ridge, eastern Tennessee. The second is the oak-
hickory forest in Ozark, central Missouri (MO) within
10 miles of the Missouri River. Two flux towers are
presently operating in the Oak Ridge forest at an
elevation around 335 m above sea level. One is called
the Walker Branch (WB) tower, which has been in
operation continuously since 1997, and stands 40 m tall
on one of East Tennessees long parallel ridges
(355703600 N, 841702400 W). The second tower, calledthe Chestnut Ridge (CH) tower (355504800 N,841904900 W), came into operation in 2005, is 60 mtall and is located on the same ridge 5 km south west of
WB. The MO site has been operating since 2004 with
one 30 m flux tower on the Ozark highland at an
elevation of 258 m above sea level.
While the Oak Ridge forest underwent some tree
damage several years ago due to bark beetle infections,
it remains fairly dense averaging about 25 m tall, while
the forest stands in the Ozark average about 17 m tall.
Both forest sites often develop dense, broad-leaf
canopies during summer and then shed their leaves
completely during the end of the fall season. The
Table 1
List of site locations and vegetation types
Site name State Latitude
(deg-min-s)
Long
(deg-m
Chestnut Ridge (CH) TN 355504800 8419Walker Branch (WB) TN 355703600 8417Ozarks (MO) MO 384800000 9212Black Hills (BH) SD 440900100 1033Canaan Valley (CV) WV 390303700 7925Fort Peck (FP) MT 481803600 1050Goodwin Creek (GC) MS 341500000 8952Sioux Falls (SF) SD 442100200 9650Audubon Ranch (AG) AZ 323502500 1103Bondville (BP) IL 400003300 8817Bondville (BV) IL 400002100 8817climate is temperate with long hot-humid summers and
short mild winters. TheWB and CH sites receive around
1300 mm of annual precipitation, with monthly totals of
about 87 mm from August to October and 115 mm
from November to July. At MO, the monthly
precipitation averages about 50 mm from December
to February and 90 mm from March to November, with
about 1000 mm per year. Scattered small trees, shrubs
and herbs grow beneath the forest understories. The
soils around the towers are well-drained, with textures
ranging from silt loam to silt clay loam; and both soil
textures and soil bulk density appear to show
considerable variability attributable to surface residue
conditions and topography. Residue cover of more than
5-cm thick is present all year round on the forest floors,largely due to the slow decomposition of the copious
amount of leaf litters that normally build up during the
fall season.
2.2.2. Needle-leaf Pine forest (BH)
A single 24 m tall flux tower has been operating
continuously since 1999 in the 17 m tall Ponderosa Pine
forest located in the interior of the Black Hills, South
Dakota at about 1715 m above sea level (440900100 N,1033802400 W). The climate is temperate with long coldwinters and short warm summers. The annual pre-
cipitation averages about 550 mm, with monthly totals
around 65 mm from April to October and 23 mm from
November to March. The evergreen pine forest is the
predominant species in this area of the Black Hills.
Except for scattered, small herbs and shrubs on the
forest floor, the lower-level of the forest is characterized
by large open space, as the forest canopy is concentrated
in the upper level of the forest. Selective cutting is used
to control the forest stand density, a management that
has created large open gaps in the forest canopy. The
predominant soils in the area consist largely of sandy
)
Elevation
(m)
Dominant
soils
Vegetation
336 Silty Clay Loam Deciduous forest
335 Silty Clay Loam Deciduous forest
258 Silt Loam Deciduous forest
1715 Clay Loam Pine needles
997 Loam Grassland
634 Clay Loam Grassland
105 Silt Loam Grassland
505 Clay Loam Grassland
854 Sandy loam Desert grassland
225 Silt Loam Corn/Soybean
220 Silt Loam Corn/Soybeanloam, and soil cores taken in the vicinity of the tower
show the soil profiles average moderately about 25 cm
deep, underlain by deep zones of granite bedrock,
apparently fractured with joints and fissures that permit
well-drained soils and deep root penetration for the
forest. The soil surface below the forest is completely
covered with a layer of dead plant materials about
10 cm thick composed largely of pine needles and seeds
that have for years accumulated and decomposed slowly
on the forest floor.
2.2.3. Grasslands
Five respective 4 m flux towers are used to monitor
five different grassland sites. As listed in Table 1, the
grassland sites include: (1) the desert grassland located
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179164on the Audubon Research Ranch (AG) in southwestern
Arizona (323502500N, 1103002600W); (2) the temperatewetted grassland located in Goodwin Creek (GC)
Watershed in Mississippi (341500000 N, 895201200 W);(3) the temperate grassland in the Canaan Valley (CV)
National Wildlife Refuge in West Virginia (390303700
N, 792501200 W); (4) the northern plain grassland in theFort Peck (FP) Reservation in northeastern Montana
(481803600 N, 1050600000 W); (5) the grassland in theBrookings (SF) Rangeland in Sioux Falls, South Dakota
(442100200 N, 965002400 W). These sites show somedistinctions attributable to grass types, regions, climate
and management practices.
The tower located at the desert grassland site in the
semi-desert plateaus of southeastern AZ was estab-
lished in 2000. It is located at an elevation of about
854 m, with scattered short grasses and shrubs. The
soils are reddish-brown, friable, deep upland sandy
loams. The climate on the desert grassland is dominated
by conditions of low precipitation cycles, high
temperature, low humidity, and gusty winds. Precipita-
tion averages about 400 mm per year, with monthly
approximations of 10 mm from January to March,
28 mm from April to June, and 48 mm from July to
December, making the grassland growth almost entirely
dependent on the precipitation cycles.
The grassland sites of FP, SF and CV undergo
seasonal winter conditions. The FP site in northeastern
Montana has been operating since 2000. It is 48 km east
of Wolf Point, MT, about 72 km from the Canadian
border, and located in a grazing area of the Great
Rolling Plains Grasslands on notably flat topography at
634 m above sea level. The soils consist of deep,
moderately drained clay loams and the soil surface
shows evidence of a large excess of precipitated calcium
carbonate. The average annual precipitation is about
100 mm during winter months fromNovember to April,
and roughly 200 mm rainfall is received during the
growing season, which is freeze-free and often runs
from May to September. The grassland site of SF was
installed in 2004 within 6 km northwest of Brookings,
SD. The topography is gentle rolling plains at 505 m
above sea level with deep, well drained clay loam soils.
The site receives about 150 mm of precipitation during
winter months between November and April, while
about 400 mm of rainfall is received during the growing
season from May to September with the freeze-free
period is slightly longer than exists for theMontana site.
The CV grassland site was also installed in 2004. It is
located in the mountain valley region at 997 m above sea
level consisting of deep loamy soils. This area receives
annual precipitation of about 500 mm during the wintermonths from November to March and 800 mm of
rainfall during freeze-free period often running from
April to October, making the growing season at the CV
site longer than the grasslands of SF and FP.
The GC site was installed in 2002. This site is
situated on gentle sloping rolling plains at elevation of
105 m. The soils are predominantly deep, well drained
silt loam textures. Covered with short grasses and
scattered trees and shrubs, this site is used for grazing.
As Mississippi winters are not very cold, snows may
occur in January and February, the growing season runs
from March to October, and the average annual rainfall
is over 1300 mm.
2.2.4. Croplands
The crop sites consist of two adjoining non-irrigated
soybean and corn fields, each with a 10 m tower, located
on the plains of Champaign in eastern Illinois around
220 m elevation. The towers are separated by a north
south distance of about 350 m. The south tower
(400002100 N, 881702400 W), denoted as BV, has beenrunning since 1997 and the north tower (400003300 N,881702600 W), denoted as BP, was installed in 2003; theBP tower is operated in collaboration with the Illinois
State Water Survey, Department of Natural Resources,
University of Illinois at Champaign/Urbana. The crops
are produced in yearly rotation (soybean is produced in
even years and corn is produced in odd years) on no-till
soils with crop residues left on the soil surface. The area
is largely on flat plains with deep, moderately well
drained silt loam soils, receives annual precipitation of
about 900 mm with a good amount of snow accumula-
tion during winters, and the summers are often hot and
humid with large rainfall.
2.3. Data analysis
The radiation data used for this study consisted of the
30-minmeasurements of the incoming and reflected solar
radiation and PAR above the vegetation canopy at the 11
flux towers. The dataset is composed of continuous
measurements throughout the year. The incoming and
reflected PAR components were measured using two
Quantum sensors (Apogee Instruments, Inc., Logan,
UT), while the solar radiation components were
measured by the CNR-1 Net Radiometer (Kipp &
Zonen, Delft, The Netherlands) with two pyranometers,
one facing upward for measuring incoming solar
radiation, and the other facing downward for measuring
reflected solar radiation from the surface. TheCNR-1Net
Radiometer is also designed with two pyrgeometers to
measure the incoming and reflected thermal radiation. In
-
order to avoid shading effects of the instruments on the
surface and to enhance spatial averaging of the radiation
measurement, the sensors were mounted horizontally on
one end of a 4 m aluminum boom with its other end
mounted south of the flux tower at heights greater than
2 m above the vegetation canopy. The field-of-view of
the sensors is around 150 which provide spatialaveraging of measurements over a circular area with a
radius of about 10H (where H is the height of the sensorabove the surface). Fig. 1 shows an exampleof theCNR-1
Net Radiometer mounted on a 10 m flux tower above the
canopy of corn crops at the BV site. The signals from the
sensors were recorded using data loggers and 30-min
averages of the radiation data were recorded in a
computer database for later analysis.
The daytime radiation flux data taken between 10:00
and 14:00 h during non-precipitation conditions were
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 165used in the calculation of RNIR and RVIS as given byEqs. (1) and (6). Values of RNIR and RVIS were used tocalculate NDVI in Eq. (7), and the LAI was estimated as
the exponential function in Eq. (8). The daily values of
RNIR, RVIS, and NDVI were calculated by averagingtheir 30-min values between 10:00 and 14:00 h.
Estimates of LAI were evaluated against field measure-
ments from the tower sites in the corn and soybean crops
at BP and BV. These direct field measurements were
provided for 20012005 by Dr. Carl Bernacchi,
Assistant Professor at the Illinois State Water Survey,
University of Illinois at Champaign/Urbana.
The tower-derived NDVI and LAI were compared
with 16-day NDVI and 8-day LAI products from the
MODIS ASCII subset database that was generated for
7 km7 km areas centered on the flux towers. Detaileddescriptions of the MODIS ASCII subset data for flux
Fig. 1. The CNR-1 Net Radiometer mounted on a 10 m tower above a
corn canopy, Bondville, IL.tower locations around the world have been provided
by ORNL DAAC (see http://www.modis.ornl.gov/
MODIS/modis.html). These MODIS ASCII subset data
for 7 km 7 km areas were derived from the standard1200 km 1200 km MODIS products. The subsettingwas done by the MODIS Science Data Support Team
(MODAPS) at NASAs Goddard Space Flight Center.
To generate the subset data, they first subsetted the
1200 km 1200 km MODIS tiles to 11 km 31 kmarrays, which were then subsetted to 7 km 7 km areasand reformatted from HDF-EOS to ASCII subsets using
an improved version of the MODIS Reprojection Tool
(MRT) in combination with code developed at the Oak
Ridge National Laboratory Distribution Active Archive
Center (ORNL DAAC). The Quality Analysis (QA) of
the ASCII subset data was based on the main algorithm
used by MODAPS along with a more robust method to
select quality control criteria. In this study, the MODIS
ASCII subset data consisted of NDVI and LAI which
were extracted as 3 km 3 km pixels at the center of the7 km 7 km areas in order to help reduce the potentialfor geolocation errors and ensure the ASCII subsets were
representative of our flux tower sites. The quality control
flags generated for the subset data were used to assess the
validity of both the NDVI and LAI products.
To ensure consistency in the tower radiation data
used to calculate the tower NDVI using Eq. (7), the
datasets between 10:00 and 14:00 h were carefully
screened to remove extreme cloudy conditions and
spurious data points. Extreme cloudy conditions were
assumed as periods when incoming solar radiation was
less than 300 W m2 and/or precipitation was greaterthan 0. This issue of bad data points was particularly
important for snowy conditions during winters as snow
cover resulted in very large albedo values. Missing
values in the radiation data were not gap-filled for the
analysis. The careful analysis of NDVI over multiple
years resulted in the maximum and minimum NDVI
values used in the LAI calculations for each site
(Table 2). In the LAI calculations, NDVI values that
occurred outside the minimum-to-maximum range
listed in Table 2 were eliminated as outliers due to
extremely distorted values of RNIR and RVIS. Whilenegative NDVI is often observed in raw remote sensing
satellite data (before the atmospheric corrections for the
VIS radiation has been applied) (Campbell and Nor-
man, 1998), the NDVI values estimated in this study
were usually greater than zero during vegetation dry-
down conditions, and RNIR and RVIS often exceed zerofor large bare soil conditions. In addition, daily values
of the tower-derived data were aggregated into 8- and
16-day averages, which reduced the large fluctuations in
-
the daily values of NDVI and LAI, and provided data on
the same time-scales as the MODIS satellite data.
3. Results
3.1. Partitioning solar radiation into wavebands
The partitioning of incoming solar radiation into the
PAR, visible, and near-infrared wavebands from
measurements at three co-located GEWEX and Sur-
face Radiation (SURFRAD) tower sites showed
fractions around 0.35 in the PAR, 0.56 in the near-
infrared (NIR), and 0.45 in the visible (VIS) wave-
bands, which were close to values reported by Weiss
and Norman (1985). The respective fractions showed
nearly constant annual values during the growing
season with acceptable daily scatter at and across sites
(Fig. 2). This means there was little variation about the
daily averages of the fractions of PAR, VIS, and NIR
during the growing season. Comparison between the
waveband fractions at the flux sites for 2004 showed a
small but clear variation in the fractions of solar
radiation in the NIR and VIS waveband at FP, where
fractions for the NIR waveband increased slowly from
about 0.52 during the beginning and end of the year to
peak values around 0.58 during the middle of year; at
the same time, the fractions for VIS waveband
decreased slowly from around 0.5 during the beginning
and end of the year to minimum values around 0.4
during the middle of year. This result was typical of all
the other years. At the BV and GC sites, the NIR and
VIS fractions showed scatter in daily values with no
clear difference in patterns of variations.
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179166
Table 2
The site-based maximum and minimum NDVI (NDVImax and
NDVImax ) and scaling constant (K) used to calculate LAI from Eq. (8)
Site State NDVImax NDVImin K
Deciduous (CH) TN 0.90 0.30 4.30
Deciduous (WB) TN 0.90 0.30 4.30
Deciduous (MO) MO 0.90 0.30 4.30
Pine forest (BH) SD 0.70 0.30 2.50
Grassland (CV WV 0.85 0.25 1.50
Grassland (FP) MT 0.80 0.25 1.70
Grassland (GC) MS 0.80 0.25 1.80
Grassland (SF) SD 0.80 0.30 1.00
Desert grass (AG) AZ 0.60 0.15 0.70
Cropland (BP) IL 0.90 0.30 2.40
Cropland (BV) IL 0.90 0.25 2.40Fig. 2. The fractions of incoming radiation in the wavebands of near-infrared
PAR and solar radiation fromGEWEX sites, during 2004, where the SURFRA(NIR) and visible (VIS) from Surface Radiation (SURFRAD) sites, and
D and GEWEX towers were co-located at BV, FP and GC (see Table 1).
-
The reflectivity of PAR, solar radiation, and visible
wavebands based on the field measurements of the
respective incoming and outgoing components showed
temporal variations of the surface conditions at the tower
sites (Fig. 2). Daily reflections of the VIS radiation were
greater than that of the solar radiation which was slightly
greater than that of the PAR radiation. The reflectivity
showed large scatter during thewinter season for the sites
at BVand FP, where, for example, the reflection values of
theVIS radiationwrongly exceeded 1 due largely to snow
buildup on the surface of the upward facing radiation
sensor. Thus, reflection values greater than 1 were not
used in the calculation of NDVI. The normal range of
reflectivity values was 0.30.6 for the VIS, 0.030.15 for
the PAR, and 0.10.3 for solar radiation during non-
snowy conditions, with reflections lower during active
vegetation growth in the summer season than the periods
of low vegetation during the spring and fall.
3.2. Intra-annual variation of the vegetation indices
at flux towers
Fig. 3 shows the time series of the daily NDVI,
reflectance in the visible (RVIS), and reflectance in
near-infrared (RNIR) wavebands for 2005 as derivedfrom radiation flux measurements at the 11 tower sites
in crops, grasslands, and forests. This year was
selected because it contained the best data sets for
all the sites. The NDVI clearly showed temporal
variation of the vegetation conditions at the various
flux tower sites as unique daily vegetation spectral
patterns for the different types of vegetation. For the
deciduous forests, NDVI increased rapidly with the
regrowth of new leaves in the spring, peaked to about
0.8 during early summer, and decreased gradually after
mid-summer before decreasing steadily during the end
of fall as the forests shed their leaves. During the same
time, RNIR closely followed the NDVI reaching amaximum value of about 0.3, while RVIS showed thereverse decreasing to a minimum value of about 0.03.
Unlike the deciduous forests, the pine forest showed
constant vegetation spectral indices of around 0.6 for
NDVI, 0.25 for RNIR, and 0.05 for RVIS during non-winter conditions. For the crops, NDVI and RNIRresembled a bell-curve with peak values of about 0.9
and 0.4, respectively, during the summer growing
season while RVIS decreased to minimum values ofabout 0.04.
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 167
cted NFig. 3. Daily values of tower-derived NDVI and refle IR and VIS for the various tower sites during 2005.
-
For the grasslands, the dissimilar growing conditions
between sites were reasonably approximated in the
distribution of vegetation spectral indices of the sites
during the year. The vegetation spectral indices were
similar for the plain grasslands at FP and SF. At both
sites, NDVI increased slowly in the spring to maximum
values of 0.6 during summer and then decreased
gradually from mid-summer to the end of fall, while
RNIR and RVIS decreased slowly to minimum values of0.25 and 0.05, respectively, during the summer.
Similarly, the NDVI showed maximum values of 0.8
for the grassland at CV, and a maximum value of about
0.6 for the grassland site at GC. For CV, the large scatter
at the beginning of the growing season was due to
surface snow conditions and the gap in the NDVI profile
was the result of missing data due to sensor failures
during 2005. On average from 2002 to 2005 the NDVI
indicated longer growing seasons for the grasslands at
CV and GC than for the two grasslands FP and SF
(Fig. 4). Of particular interest is the double peak in the
NDVI curve at GC during 2005 (Fig. 3), which occurred
because of the inadvertent use of weed killers on the
grass around the flux towers in June and August. For the
desert grassland at AG, the annual pattern of NDVI was
strongly related to the precipitation cycles.
One of the primary reasons for this study was to
evaluate the feasibility of using routine tower measure-
ments to derive vegetation indices that can serve as an
adequate description of vegetation cover such as LAI.
Daily vegetation indices were derived using 30-min
tower measurements of solar and PAR radiation fluxes
above the vegetation canopy between hours 10:00 and
14:00 during non-precipitation conditions. However,
measurement of the radiance reflected from the surface
is dependent upon the irradiance intensity arriving at the
surface and the position of the sensor above the surface.
The irradiance intensity can vary with the sun position
high at around noon and low in the mornings and
evening and the wavebands. This means the measured
radiance reflected from the surface also changes as a
function of the position of the sensor relative to the sun,
and the wavelength. Thus, for the same cloud
conditions, fixed tower sensors looking at the ground
surface in the mornings and evenings would record low
reflectance while high reflectance would be measured at
noon at the same spot. The effect of the sensor position
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179168
s uniFig. 4. Daily values of tower-derived NDVI show annual distribution que to vegetation types at the various tower sites during 20022005.
-
is often described in terms of the surface bidirectional
reflectance distribution function (BRDF), which is
defined as the ratio of the radiance reflected from the
surface into the direction of the sensor to the incident
irradiance illuminating the surface from a source
direction (Disney et al., 2004). Without BRDF
corrections discrepancies may result in determining
changes of the surface conditions using measurements
of radiance reflected from the surface.
To evaluate the likely variations of the vegetation
indices between hours 10:00 and 14:00, the standard
deviations were computed for the daily values of RNIR,RVIS and NDVI derived as the average of the 30-minvalues between hours 10:00 and 14:00. The summary of
these statistics computed for DOY 120300 is listed in
Table 3, with the means of incoming and outgoing solar
and PAR radiation for each measurement site. Overall,
there was very little variation in the 30-min values of
the vegetation indices between hours 10:00 and 14:00,
as RNIR averaged 0:241 0:011, RVIS averaged0:067 0:002, and NDVI averaged 0:568 0:014.These small standard deviation values made it appro-
priate to derive daily values RNIR, RVIS and NDVI byaveraging their 30-min values. In addition, the small
standard deviation values also suggested that BRDF
corrections were unnecessary and the Lambertian
reflectance may have dominated the 30-min measure-
ments between hours 10:00 and 14:00 that were
considered for this study.
3.3. Inter-annual variation of NDVI
Fig. 4 shows the annual course of the tower-derived
NDVI from 2002 to 2005 for the 11 tower sites. The
NDVI showed clear variation during the year with
maxima during the summer season and minimum values
during early spring and winter seasons. Except for the
pine forest, all the sites showed significant NDVI peak
during the summer; NDVI for the pine site remained near
0.5 for the whole year; the observed decrease and large
scatter during winter were related to the snow cover on
the ground/canopy. The deciduous forests showed a
broader distribution centered on summer with true
maximumNDVI around 0.8 andminimumvalues around
0.4 during winter; the maximum NDVI in the desert
grassland was 0.5, much less than the forests. For the
crops, NDVI showed a Gaussian distribution centered on
mid-summer with maximum NDVI around 0.85 and a
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 169
Table 3
Annual averages of the daily incoming solar (Solrdn) and PAR (PARdn) radiation and outgoing solar (Solrup) and PAR (PARup), the daily derived
as the
DVI)
4:00 h
)
P
(
m
n
2
7 4
7 3
3 5
1 7
1 6
2 6
1 2
7
1 7
4 4
3
3 4
8 4
6
3 8
2 1
1 6
9 1
5 9
4 7near-infrared (RNIR) and visible (RVIS) wavebands, and NDVI, as well
near-infrared (stdnir) and visible (stdvis) wavebands, and NDVI (stdN
Table 1; daily values were based on data taken between 10:00 and 1
Year Site State Solrdn
(W m2)Solrup
(W m2)PARdn
(mmol
m2 s1
2004 Deciduous (CH) TN n/a n/a n/a
Deciduous (WB) TN 539.71 81.97 845.75
Deciduous (MO) MO 619.95 71.16 1136.3
Pine forest (BH) SD 605.70 49.62 1052.1
Grassland (CV) WV 581.53 109.50 1120.2
Grassland (FP) MT 572.92 89.88 1008.4
Grassland (GC) MS 639.81 109.23 1032.0
Grassland (SF) SD 610.83 117.16 1078.9
Desertgrass (AG) AZ 815.54 159.04 1544.2
Cropland (BP) IL 389.03 75.33 700.15
Cropland (BV) IL 637.48 138.04 1021.1
2005 Deciduous (CH) TN 573.70 83.62 1064.8
Deciduous (WB) TN 582.57 79.60 954.72
Deciduous (MO) MO 711.34 84.95 1343.3
Pine forest (BH) SD 625.27 53.05 1053.4
Grassland (CV) WV 586.29 107.41 917.94
Grassland (FP) MT 599.43 86.90 1052.9
Grassland (GC) MS 770.77 130.98 1204.2
Grassland (SF) SD 640.78 112.73 1058.9
Desertgrass (AG) AZ 812.25 155.59 1507.4
Cropland (BP) IL 607.06 125.26 1218.8
Cropland (BV) IL 601.32 106.95 1148.1daily standard deviations in using 30-min values of reflectances in the
during 2004 and 2005 for DOY 120300 at the flux towers listed in
ARup
mmol2 s1)
RNIR stdRnir RVIS stdRvis NDVI stdNDVI
/a n/a n/a n/a n/a n/a n/a
9.22 0.247 0.047 0.035 0.0013 0.721 0.0228
1.10 0.180 0.006 0.037 0.0018 0.658 0.0129
9.44 0.119 0.0043 0.038 0.0009 0.518 0.0142
6.23 0.303 0.012 0.050 0.0020 0.721 0.0128
6.44 0.226 0.009 0.078 0.0017 0.495 0.0117
5.86 0.259 0.010 0.065 0.0033 0.597 0.0133
2.21 0.302 0.008 0.060 0.0016 0.668 0.0049
10.08 0.241 0.0031 0.136 0.0022 0.283 0.0087
8.03 0.232 0.024 0.114 0.0041 0.362 0.0348
8.10 0.332 0.010 0.077 0.0028 0.607 0.0095
7.72 0.230 0.008 0.046 0.0024 0.666 0.0132
1.51 0.222 0.007 0.034 0.0021 0.734 0.0132
8.88 0.187 0.006 0.037 0.0018 0.669 0.0140
3.35 0.122 0.005 0.041 0.0016 0.512 0.0147
0.23 0.283 0.008 0.057 0.0015 0.668 0.0090
3.65 0.203 0.009 0.083 0.0025 0.424 0.0106
15.96 0.266 0.007 0.097 0.0023 0.465 0.0158
7.28 0.270 0.009 0.066 0.0025 0.612 0.0066
71.98 0.254 0.004 0.114 0.0019 0.381 0.0091
3.34 0.309 0.030 0.081 0.0021 0.568 0.0313
3.05 0.271 0.010 0.066 0.0017 0.607 0.0069
-
minimum around 0.35. For the other temperate grass-
lands, NDVI showed various distributions across sites as
maximum values ranged from 0.6 to 0.8 and minimum
values were about 0.25. However, the maximum NDVI
may have suffered fromNDVI saturation in the crops and
in the deciduous forests, because their NDVI values
tended to plateau during summer, even though LAI may
have continued to increase.
The magnitudes of NDVI also displayed slight
year-to-year differences at each of the tower sites. For
the temperate grasslands, the peak values of NDVI
during 2004 slightly exceeded those during 2005, as
the onset and rate of both green-up and dry-down
varied slightly between years; the magnitudes of NDVI
from 2002 to 2005 displayed the most dynamic
variations at the FP and GC sites. For the desert
grassland/shrub at AG, the peak values of NDVI
during 2002 and 2005 slightly exceeded those during
2003 and 2004. The yearly datasets were inadequate to
show year-to-year variations at the deciduous forest
sites at MO and the one at CH. However, for WB,
while the peak values of NDVI showed only a small
variation from 2002 to 2005, the annual distribution of
NDVI showed clear differences corresponding to the
variation in the onset and rate of green-up as well as
senescence from year-to-year. Similar year-to-year
patterns of NDVI were also displayed for the crops at
BVand BP. For the pine forest and BH, the magnitudes
of NDVI from year-to-year were nearly constant
during throughout the summer, but showed large
scatter during the winter.
3.4. Comparison of NDVI
Fig. 5 shows an example of the annual variation of
the tower-derived NDVI along side the MODIS NDVI
during 2004 for 10 flux tower sites. The MODIS NDVI
was a normalized ratio of the near-infrared (841
876 nm) and red (620670 nm) reflectance, produced at
16-day intervals, and was extracted as 3-km subpixels at
the center of the 7-km arrays centered on the flux
towers. The daily tower-derived NDVI was aggregated
into 16-day values for direct comparison with the
MODIS product. In general, NDVI values fromMODIS
exceeded those from the tower for the forests, but the
tower-derived and MODIS NDVI values were much
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179170
I fromFig. 5. Annual variation of 16-day averages of NDV flux towers and the MODIS satellite during 2004.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 171closer for the grasslands and crops. Regression of the
tower-derived NDVI against the MODIS NDVI showed
large scatter for all the sites (Fig. 6). The slopes and
root-mean-square errors (RMSE) for the linear fits were
calculated to quantify the comparison (Table 4). Even
though Fig. 6 shows considerable scatter, the overall
agreement for most of the sites was good between the
tower-derived NDVI and MODIS. Slopes of the straight
lines with zero intercept between the tower-derived and
MODIS NDVI ranged from 0.72 to 1.18 with RMSE
values of 0.090.20. Values of the correlation coeffi-
cient (R) also varied with sites: for the desert grassland
at AG, R 0:53; and for the pine forest FP, R 0:57which were lower than in the temperate grasslands at the
CV, FP and GC sites where R averaged about 0.79 and in
the crops at BV and deciduous forest sites where R
averaged about 0.87.
3.5. Leaf area index estimated from tower-derived
NDVI
The daily leaf area index (LAI) of the vegetation at
the flux tower sites was calculated by fitting an
Fig. 6. Comparison of tower-derived NDVI with 16-day averages from thexponential function to the derived-NDVI described
in Eq. (8). Table 2 shows the maximum and minimum
NDVI, and scaling constants for the 11 flux tower sites.
The range of NDVI was determined by analyzing the
seasonal variations of NDVI from 2002 to 2005, and the
scaling constant (K) was selected so that the exponential
function prediction of maximum LAI approximated the
nominal maximum LAI values for the vegetation types
at the flux towers.
The use of the exponential function relating LAI and
NDVI estimated daily LAI with annual distributions
similar to the NDVI during the year. The periods of
maximum and minimum LAI correlated directly with
those of NDVI (Fig. 7). Like the NDVI, the daily tower-
derived LAI showed increases from minimum values in
spring, maximum during summer and minima again
during fall and winter, similar to the distribution of
NDVI shown in Fig. 7. Magnitudes of LAI during the
year showed differences relative to the crop, grassland,
and forest sites where the maximum LAI ranged from
the highest values of about 7 m2 m2for the deciduousforests to the smallest values of 0.6 m2 m2for thedesert grassland.
e MODIS satellite at the various GEWEX sites during 20032005.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179172Table 4
Statistics of the comparisona of tower-derived NDVI with MODIS
NDVI in 16-day averages from 2003 to 2005 for the GEWEX sites
Site State N Slope R RMSE
NDVI
Deciduous (CH) TN 15 0.80 0.94 0.16
Deciduous (WB) TN 60 0.9 0.84 0.11
Deciduous (MO) MO 31 0.83 0.92 0.14
Pine forest (BH) SD 53 0.73 0.57 0.20
Grassland (CV WV 30 0.90 0.77 0.13
Grassland (FP) MT 31 1.15 0.84 0.10
Grassland (GC) MS 55 0.77 0.75 0.17
Grassland (SF) SD n/a n/a n/a n/a
Desertgrass (AG) AZ 56 1.18 0.53 0.09
Cropland (BP) IL n/a n/a n/a n/a
Cropland (BV) IL 43 1.01 0.87 0.13
LAI
Deciduous (CH) TN 26 0.71 0.88 1.74
Deciduous (WB) TN 93 1.02 0.95 0.76
Deciduous (MO) MO 57 0.77 0.95 1.17
Pine forest (BH) SD 88 0.61 0.66 1.37
Grassland (CV WV 53 0.85 0.77 0.71
Grassland (FP) MT 90 1.15 0.75 0.33
Grassland (GC) MS 89 0.70 0.68 0.68
Grassland (SF) SD n/a n/a n/a n/a3.6. Comparison of LAI
Comparisons between the 8-day averages of tower-
derived LAI and the 8-day MODIS LAI showed good
annual distributions at the nine flux towers where
MODIS data were also available (Fig. 8); similarly, the
scatter plots in Fig. 9 show good statistical agreement
between the tower-derived LAI and MODIS LAI.
Table 4 shows the statistics of the 1:1 regression lines
with zero intercept of the tower-derived LAI versus
MODIS LAI for all the sites, where the slopes ranged
from 0.60 to 1.28, RMSE from 0.19 to 1.74, and R from
0.68 to 0.95. The largest difference in the comparison
occurred over the pine forest with slope of 0.61 and R of
0.66, indicating MODIS LAI values exceeded tower-
derived LAI, and the best agreement occurred over the
deciduous forest atWBwith slope of 1.02 and R of 0.95.
A comparison of both tower-derived andMODIS LAI
wasmadewith direct fieldmeasurements ofLAI to assess
the performance of the remote sensors.Wewere only able
to obtainmeasurements of LAI for the corn/soybean crop
site at BV during 20022005. The tower-derived and
Desertgrass (AG) AZ 117 0.74 0.76 0.19
Cropland (BP) IL n/a n/a n/a n/a
Cropland (BV) IL 97 1.28 0.93 0.62
a N is the number of observations, RMSE is the root-mean-square of
the difference between tower and MODIS values, Slope is based on
linear fit with zero intercept, and R is the correlation coefficient of the
regression line; NDVI data were recorded in 16-day intervals with 8-
day intervals for LAI.MODIS LAI agreed quite well with the field data for
these 4 years in the corn/soybean with MODIS LAI
slightly lower than both the measured and tower-derived
LAI (Fig. 10). For the corn, the average maximum
measured LAI was about 4.5 m2 m2compared withabout 6 m2 m2 for soybean.
4. Discussions
This study showed the performance of single point
measurements above the vegetation in deriving vegeta-
tion indices to characterize the variations of the
biophysical and structural properties of site-specific
vegetation types. The visible and near-infrared fractions
of solar radiation measured at flux towers within both
the GEWEX and SURFRAD networks were nearly
constant throughout the year; they were about 0.45 for
visible and 0.55 for near-infrared. The visible fraction
of solar radiation was very similar to values reported by
Weiss and Norman (1985). However, the PAR fraction
of solar radiation was around 0.35. This discrepancy
may have resulted from the differences in sensor
calibration and site characteristics between the
GEWEX and SURFRAD sites. The conservative ratios
for the fractions of solar radiation in the visible and the
near-infrared wavebands made it possible to success-
fully derive RNIR, RVIS, and NDVI, as well as the LAI, at11 different flux towers in various grasslands, crops, and
forests across the United States using only measure-
ments of solar radiation and PAR fluxes above the
vegetation. There were site disturbance problems that
may have contaminated RNIR, RVIS, and NDVI at BH,BV, FP, and GC during 2002 and at GC during 2005
(Figs. 3 and 4). The disturbance resulted mainly from
the installation of the flux towers and the inadvertent use
of weed killers in the vicinity of the tower at GC as
mentioned in Section 3.2. Notwithstanding the inter-
ference from the disturbance, the daily RNIR, RVIS, andNDVI closely matched the vegetation canopy greenness
where RVIS was consistently less than RNIR, and bothreflectances were much less than the NDVI. The highest
values of NDVI and the lower values of RVIS occurredduring the peak growing season at nearly all the tower
sites; however, the seasonal distribution of RNIR tendedto vary with site, suggesting that while NDVI and RVISwere sensitive to the vegetation greenness, RNIR wasmore influenced by variations of the vegetation canopy
structure and density. Daily values of RNIR mimickedthe distribution of NDVI for the crops and deciduous
forests, but showed fairly stable distribution with only a
slight decrease during the peak growing season for the
grasses and the pine forest. During the winter period
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 173from DOY 1 to 120 the large albedo of snow caused
RNIR and RVIS to increase sharply to values considerablyhigher than observed during peak growing season,
producing negative values of NDVI.
The NDVI showed clear differences in the onset,
magnitude, and duration of the vegetation greenness
that seemed to depend upon the locations, climates and
vegetation types present at the 11 tower sites. The time
series of the NDVI depicted the dynamic behavior of the
different vegetation sites with values that showed a
rapid increase during the green-ups and a gradual
decrease during the dry down periods. The magnitudes
of NDVI for all the sites ranged from 0:4 during snowconditions to 0.85 during the peak stages of vegetation
growth. The highest peak NDVI values ranged as
follows: 0.85 for the crops at BV and BP, 0.81 for the
deciduous forest at WB and CH, 0.80 for the grassland
at CV, 0.78 for the grasslands at SF, 0.75 for the
deciduous forest at MO, 0.66 for the grassland GC, 0.65
for the grassland at FP, 0.61 for the pine forest at BH,
and 0.52 for the desert grassland/shrub at AG. The
lowest NDVI values during non-snow conditions
averaged between 0.3 and 0.4. For the crops at BV
and BP, the profile of NDVI depicted a Gaussian
Fig. 7. Daily estimated LAI as an exponential function of towerdistribution of greenness with onset, peak, and start of
dry-down around DOY 150, 200, and 230, respectively.
For the deciduous forests, the NDVI distribution
produced a broad growing season that started around
DOY 90, peaked around DOY 150, and started dry
down around DOY 270. The temperate grasslands
showed different growing patterns in their seasonal
NDVI curves, largely attributable to the different
climate conditions among their locations. From the
NDVI distribution, the onset of greenness, peak
greenness, and dry-down period appeared to start
around DOY 70, 120, 270, respectively, at the grass-
lands at GC, DOY 100, 150, 250 at CV, 100, 170, 200 at
SF, and DOY 120, 150, 200 at FP. The seasonal
variation of NDVI observed over the desert grassland
site at AG strongly followed the precipitation cycles. In
the case of the evergreen pine forest at BH, NDVI
showed no strong greenness patterns with only slight
stable increase in NDVI during the non-snow periods
from DOY 120 to 300. Clearly, the periods in DOY
between the onset of greenness and end of dry-down
depicted by the seasonal distribution of NDVI
successfully indicated the length of the growing season
at the various sites. Overall, the magnitudes of the
-derived NDVI for the various tower sites for 20022005.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179174NDVI values seemed to vary depending on the
combined effects of the vegetation type, climate, and
the management strategy at the sites.
When the tower- and the MODIS-derived NDVI
were compared over 9 of the 11 sites in this study, both
data sets showed good agreement in depicting the
seasonal greenness at all the sites. The analysis of the
regression line fitted to the tower- and MODIS-derived
NDVI data sets for 20032005 showed poor agreement
for desert grassland at AG (R 0:53, slope 1:18) andthe pine forest at BH (R 0:57; slope 0:72) andslightly above average agreement for the grassland at
CV (R 0:77; slope 0:98) and at GC (R 0:75;slope 0:77), with a better fit for the other five siteswhere R> 0:80 and the slope ranged from 0.70 to 1.15,as the root-mean square error (RMSE) showed large
scatter about the 1:1 line (Table 4).
One of the primary interests of this study was to use
the in situ derived NDVI to estimate consistent values of
LAI that can account for site-specific features in support
of model simulations of energy, water, and carbon
exchange in local vegetation environments. In addition
to depicting LAI, NDVI is widely used to estimate other
biophysical parameters of the vegetation, including
Fig. 8. Annual variation of 8-day averages of LAI fromgreenness, fraction of absorbed PAR (FPAR), and
biomass (Asrar et al., 1984; Sellers, 1985; Tucker, 1979;
Lotsch et al., 2003). Empirical and synergistic algo-
rithms have been widely used to estimate LAI from
NDVI (White et al., 1997; Knyazikhin et al., 1998;
Pontailer et al., 2003; Campbell and Norman, 1998). In
this study, we used an empirical exponential function to
estimate LAI from the tower-derived NDVI over
different vegetation types. The estimated LAI provided
an adequate assessment of the vegetation cover, and
annual distributions of LAI matched fairly well with the
NDVI in depicting the periods of maximum and
minimum vegetation greenness for the 11 tower sites
located in crops, forests, and grasslands. This resulted in
the increase of LAI to peak values during summers
which decreased to minima in fall and winter for all the
sites with the exception of the pine forest at BH
which remained green throughout the year and the
desert grassland at AG which correlated with the
precipitation cycles. The maximum LAI of about
6.5 m2 m2occurred for the deciduous forests followedby 4.5 m2 m2for the crops, 24 m2 m2 for thetemperate grasses and pine forest, and 0.6 m2 m2
for the desert grassland.
flux towers and the MODIS satellite during 2004.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 175
Fig. 9. Comparison of tower-derived LAI with 8-day averages from the MODIS satellite at the various GEWEX sites during 20032005.
Fig. 10. Daily tower-derived LAI compared with LAI from direct measurements and from MODIS for the corn/soybean site (BV) in IL for 2002
2005.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179176The tower and MODIS data agreed reasonably well
in assessing the annual variations of LAI for the crop,
grassland, and forest locations that were examined in
this study. However, the 1:1 line between the two data
sets showed considerable scatter similar to the NDVI.
The regression analysis between both data sets for each
vegetation type indicated R of 0.660.95, slope of 0.61
1.28, and RMSE of 0.331.74 m2 m2 for all the sites.In particular, the tower and MODIS were quite close in
their LAI values for the deciduous forests, but the tower
had slightly higher LAI for the crops in Illinois and the
grasslands in Montana, much less LAI for the pine
forest, and slightly lower LAI over the rest of the sites.
The RMSE of the difference between the tower-derived
and MODIS LAI ranged from 0.3 to 1.74 m2 m2, andas expected, the deciduous forest (with the largest
maximum LAI) indicated the highest deviation and the
desert grassland (with the lowest maximum LAI)
showed the smallest deviation.
Because of the limited availability of field data of LAI
to provide an assessment of how well the tower-derived
LAI perform at all the sites in this study, we only
succeeded in obtaining direct LAI measurement for
20022005 at the crop site in Illinois, where soybean was
produced during 2002 and 2004 and corn during 2003
and 2005. Future work is needed to validate tower-
derived LAI against direct field measurements at all the
vegetation types that were examined in this study. In this
preliminary evaluation, the tower-derived, MODIS and
measured LAI were in good agreement with the MODIS
LAI peak lower than the nearly-identical measured and
tower-derived LAI (Fig. 10). During the soybean years,
the measured and tower-derived LAI peak reached
6 m2 m2compared with 5 m2 m2 for theMODIS LAIpeak. Similar distribution occurred during the corn years,
as the measured and tower-derived LAI peak reached
5 m2 m2 compared with 4 m2 m2 for the MODISLAI, which had larger variability and depicted less
systematic distribution during the course of the year.
Even though LAI measurements were not available
for all of the sites, the tower-derived LAI showed
distributions and peak values that were consistent with
results of similar vegetation types reported in the
literature. For instance, Hanson et al. (2003, see Table
17.2) used leaf-litter production to measure the LAI at
WB from 1993 to 2000 and reported peak values of LAI
that ranged from 5.5 to 7.0 m2 m2. Fassnacht et al.(1994) have used three different types of optical sensors
to measure LAI values of 0.53.0 m2 m2 in the openpine forest at Lubrecht Experimental Forest in MT, a
site similar to the pine forest at BH. Biomass samples
and LAI sensor were used to measure LAI valuesbetween 0.3 and 1.7 m2 m2 at a semi-arid grassland/shrub site located in the Mexican part of the Upper San
PedroRiverBasin along theMexico-USborderwhere the
annual precipitation amount and cycle were similar to
that at AZ (Nouvellon et al., 2000). Cohen et al. (2006)
used allometry, destructive harvest, and LAI-2000 sensor
to measure LAI to evaluate MODIS LAI products at
contrasting vegetation sites, and they reported measured
LAI values of about 2.03.0 m2 m2, 0.10.5 m2 m2,and 2.3 m2 m2 for Konza Prairie grassland in, shortdesert grassland in New Mexico, and pine forest in
Oregon, respectively. Frank (2003) used destructive
harvest to measure peak LAI of 0.38 and 0.44 m2 m2 ingrazed and ungrazed prairie at the Northern Great Plains
Research Laboratory, Mandan, ND, but his results were
much lower than the peak LAI estimated in our study at
the FP grazed grassland site.
Despite the encouraging performance of the tower-
derived NDVI in this study, site-specific remote
sensing techniques have had limited use in deriving
vegetation indices, especially in long-term applica-
tions, due largely to the complication of the instrument
system. As such, most remote sensing data used in
assessing vegetation conditions are obtained from
regional- and global-scale atmospheric satellite
observations (Cihlar et al., 1997; Huete et al., 2002;
Lotsch et al., 2003; Spanner et al., 1990; Sakai et al.,
1997; Zhang et al., 2004). A few short-term studies
have been reported mainly for single vegetation types.
For example, Pontailer et al. (2003) reported estimates
of LAI for 0.6 m tall scrub oaks by fitting a
exponential function to NDVI obtained from the
measurements of reflected red and near-infrared
wavebands with sensors at 0.5 m above the vegetation.
However, the study was conducted only for the months
of February and June in 2000 over 4 m2 plots. Using
hyperspectral remote sensing data from aircraft
measurements at 2500 m altitude at 5-m ground
resolution during the summer of 1999, Pacheco et al.
(2001) reported good results of LAI over several fields
of corn and bean in South Western Ontario, Canada,
but their results were poor for the individual fields.
Aircraft measurements at high-ground resolution
1.5 m were also used during the soil moisture
experiment during the summer of 2002 (SMEX02)
to derive good estimates of the LAI variability in corn
and soybean fields in the Walnut Creek Watershed
located near Ames, Iowa (Anderson et al., 2004).
Unfortunately, aircraft measurements are too expen-
sive for consistent long-term applications.
While the continuous point measurements conducted
at the 11 flux tower locations in this study provided a
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 177practical method for deriving multiple years of site-
specific annual NDVI distributions, problems of cloud
cover, sensor view/position, and surface conditions can
complicate validation of NDVI. Unlike the satellite-
derived NDVI, the tower-derived NDVI was less
affected by cloudy conditions. While theoretical
algorithms are often used to filter out erroneous cloudy
data points from satellite products (Lotsch et al., 2003;
Zhang et al., 2004), the continuous 30-min measure-
ments of precipitation and incoming solar radiation at
the flux towers made it possible to directly identify and
remove cloudy data points in the derivation of NDVI.
There was very little variation in the magnitudes of the
vegetation spectral indices with sensor viewing angle
above the canopy. The standard deviations of the daily
values of RNIR, RVIS, and NDVI derived from 30-mindata points between 10:00 and 14:00 local time
indicated variations less than 2% (Table 2). This small
variation made BRDF corrections unnecessary for the
tower-derived NDVI, similar to variations of albedo
over agricultural crops based on point measurements
reported by (Disney et al., 2004). By contrast, satellite
products often depend on BRDF and corrections
depend on the type of BRDF models used Disney et al.
(2004). In addition, downscaling large-scale satellite
products and upscaling point measurements remains a
difficult research issue to validate (Anderson et al.,
2003; Reynolds et al., 1993), and these unsolved
problem may have caused some of the differences in
the comparison between the MODIS products derived
at 1 km resolution and tower-derived NDVI/LAI.
Finally, NDVI saturation is an important issue in
relating LAI to NDVI for high density vegetation. The
NDVI saturation defines the upper limit of the
magnitude of vegetation density above which RNIRand RVIS are insensitive to farther increases ofvegetation density. The tower-derived NDVI was not
corrected for effects of the large soil component in
sparse vegetation and spectral saturation in dense
vegetation (Richardson andWiegand, 1977). However,
the MODIS LAI showed underestimation for the corn/
soybean crops in Illinois which may have resulted from
NDVI saturation, compared with the tower-derived
LAI which was nearly identical the direct LAI
measurements.
5. Conclusions
The continuousmeasurements of hourly global solar
radiation and PAR from fixed points on various flux
towers above the plant canopies reasonably produced
the annual variability of vegetation indices for grass-lands, forests, and crops. During precipitation and
winter-snowy conditions, above canopy radiation flux
measurements tended to misrepresent the vegetation
indices. This discrepancy results from the problems of
cloud cover, sensor view/position, small radiation
fluxes, and large surface snow conditions. Improve-
ment is therefore needed in surface point observations
when radiation fluxes are low and radiation scattering is
high. With the low radiation and snowy conditions
neglected in this study, our tower-derived reflectances
in the visible and near-infrared wavebands offered
encouragement that above canopy tower observations
can be used to determine vegetation indices during the
growing season. Results of this study illustrate the
usefulness of flux tower measurements for describing
the relation between radiation fluxes and the greenness
in vegetation canopies. In addition, our tower-derived
NDVI indicated that vegetation indices derived from
large-scale satellites, such asMODIS, might be limited
by the apparent role of local climate and soil-canopy
conditions. A relatively late infield snow-melt follow-
ing thewinter season can apparently delay the green-up
of a local vegetation by few weeks, compared to wet/
warm conditions. In addition, the prevalence of dry
soils in the desert due to limited rainfall provides for
scattered patches of vegetation greenness interspersed
within large swath of bare soils. Implications of these
factors must be judged within the context of models of
the soilplantatmosphere system. Flux towers are
increasingly performing well in different landvegeta-
tion systems because of the rapid advances in
instrumentation. This study suggests that producing
consistent vegetation indices from tower measure-
ments holds great promise to monitor, quantify, and
investigate changes in vegetation systems that may
validate and complement satellite data at field scales
and support the effort formodeling the seasonal profiles
of energy, water, and carbon fluxes in vegetation
environments.
Acknowledgments
This work was funded by the NASA Terrestrial
Hydrology Program under project no. PR42000-54267
as contribution to the GEWEX America Prediction
Project (GAPP). Portion of the data for this work was
provided by Dr. Carl Bernacchi, Assistant Professor at
the Illinois State Water Survey, University of Illinois
at Champaign/Urbana as well as the Moderate
Resolution Imaging Spectroradiometer (MODIS)
ASCII Subsets, Oak Ridge National Laboratory,
DAAC.
-
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179178References
Anderson, M., Kustas, W., Norman, J., 2003. Upscaling and down-
scaling-a regional view of the soilplantatmosphere continuum.
Agron. J. 95, 14081423.
Anderson, M., Neale, C., Li, F., Norman, J., Kustas, W., Jayanthi, H.,
Chavez, J., 2004. Upscaling ground observations of vegetation
water content, canopy height, and leaf area index during
SMEX02 using aircraft and landsat imagery. Rem. Sens.
Environ. 92, 447464.
Asrar, G., Fuchs, M., Kanemasu, E., Hatfield, J., 1984. Estimating
absorbed photosynthetic radiation and leaf area index from spec-
tral reflectance in wheat. Agron. J. 76, 300306.
Berk, A., Bernstein, L., Anderson, G., Acharya, P., Roberston, D.,
Chetwynd, J., Alder-Golden, S., 1998. MODTRAN cloud and
multiple scattering upgrades with application to AVIRIS. Rem.
Sens. Environ. 65, 367375.
Campbell, G., Norman, J., 1998. An Introduction to Environmental
Biophysics. Springer-Verlag, New York.
Carlson, T., Capehart, W., Gillies, R., 1995. A new look at the
simplified method for remote sensing of daily evapotranspiration.
Rem. Sens. Environ. 54, 161167.
Chen, J., Pavlic, G., Brown, L., Cihlar, L., Leblanc, S.,White, H., Hall,
R., Peddle, D., King, D., Trofymow, J., Swift, E., Van der Sanden,
J., Pellikka, P., 2002. Derivation and validation of Canada-wide
coarse-resolution leaf area index maps using high-resolution
satellite imagery and ground measurements. Rem. Sens. Environ.
80, 165184.
Cihlar, J., Chen, J., Li, Z., 1997. On the validation of satellite-derived
products for land applications. Can. J. Rem. Sens. 23, 381389.
Clevers, J., 1989. The application of a weighted infrared-red vegeta-
tion index for estimating leaf area index by correcting for soil
moisture. Rem. Sens. Environ. 29, 2537.
Cohen, W., Maiersperger, T., Turner, D., Ritts, W., Pflugmacher, D.,
Kennedy, R., Kirschbaum, A., Running, S., Costa, M., Gower, S.,
2006. MODIS land cover and lai collection 4 product quality
across nine sites in the Western Hemisphere. Trans. Geosci. Rem.
Sens. 44 (7), 18431856.
Disney, M., Lewis, P., Thackrah, G., Quaife, T., Barnsley, M., 2004.
Comparison of MODIS broadband albedo over an agricultural site
with ground measurements and values derived from earth obser-
vation data at a range of spatial scales. Int. J. Rem. Sens. 25 (23),
52975317.
Fassnacht, K., Gower, S., Norman, J., McMutrie, R., 1994. A com-
parison of optical and direct methods for estimating foliage sur-
face area index in forests. Agric. For. Met. 71, 183207.
Ferreira, L., Yoshioka, H., Huete, A., Sano, E., 2003. Seasonal
landscape and spectral vegetation dynamics in the Brazilian
Cerrado: an analysis within the Large-Scale Biosphere-Atmo-
sphere Experiment in Amazonia (LBA). Rem. Sens. Environ. 87,
534550.
Frank, A., 2003. Evapotranspiration from Northern Semiarid Grass-
lands. Agron. J. 95, 15041509.
Hall, F., Peddle, D., LeDrew, E., 1996. Remote sensing of biophysical
variables in boreal forest stands of Picea Mariana. Int. J. Remote
Sens. 17, 30773081.
Hall, F., Shimabukro, Y., Huemmrich, K., 1995. Remote sensing of
forest biophysical structure using mixture decomposition and
geometric reflectance models. Ecol. Appl. 5, 9931013.
Hall, R.J., Davidson, D.P., Peddle, D.R., 2003. Ground and remote
estimation of leaf area index in Rocky Mountain forest stands,
Kananaskis, Alberta. Can. J. Rem. Sens. 29 (3), 411427.Hanson, P., Todd, D., Joslin, J., 2003. Ecological Studies: North
American Temperate Deciduous Forest Responses to Changing
Precipitation Regimes, vol. 166. Springer, New York, Ch. Canopy
production, pp. 303315.
Huete, A., Didan, K., Miura, T., Rodriguez, E., Gao, X., Ferreira, L.,
2002. Overview of the radiometric and biophysical perforamance
of the modis vegetaion indices. Rem. Sens. Environ. 83, 195213.
Knyazikhin, Y., Martonchik, J., Diner, D., Myneni, R., Verstaete, M.,
Pinty, B., Gobron, N., 1998. Estimation of vegetation canopy leaf
area index and fraction of absorbed photosynthetically active
radiation from atmosphere-corrected MISR data. J. Geog. Res.
103 (24), 32239322256.
Li-Cor, I., 1990. LAI-2000 Plant Canopy Analyzer: Operating Man-
ual. LI-COR, Inc., Lincoln, NE.
Lotsch, A., Tian, Y., Friedl, M.A., Myneni, R., 2003. Land cover
mapping in support of LAI and FPAR retrievals from EOS-
MODIS and MISR: classification methods and sensitivities to
errors. Int. J. Rem. Sens. 24 (10), 19972016.
Moon, P., 1940. Proposed standard solar radiation curves for engi-
neering use. J. Franklin Inst. 11, 583617.
Nouvellon, Y., Begue, A., Moran, M., Seen, D., Rambal, S., Luquet,
D., Chehbouni, G., Inoue, Y., 2000. PAR extinction in shortgrass
ecosystems: effects of clumping, sky condition and soil albedo.
Agric. For. Meteorol. 105, 2141.
Pacheco, A., Bannari, A., Deguise, J., McNairn, H., Staenz, K., August
2124 2001. Application of hyperspectral remote sensing for lai
estimation in precision farming. In: Proceedings of the 23rd
Canadian Remote Sensing Symposium, Sainte-Foy, QB, Canada.
Peddle, D., Grunke, S., Hall, F., 2001. A comparison of spectral
mixture analysis and ten vegetation indices for estimating boreal
forest biophysical information from airborne data. Can. J. Rem.
Sens. 27, 627635.
Peddle, D., Hall, F., LeDrew, E., 1999. Spectral mixture analysis and
geometric optical reflectance modeling of boreal forest biophy-
sical structure. Rem. Sens. Environ. 67, 288297.
Peddle, D., Johnson, R., 2000. Spectral mixture analysis of airborne
remote sensing imagery for improved prediction of leaf area index
in mountainous terrain, Kananaskis, Alberta. Can. J. Rem. Sens.
26, 176187.
Pontailer, J., Hymus, G., Drake, B., 2003. Estimation of leaf area index
using ground-based remote sensed NDVI measurements: valida-
tion and comparison with two indirect techniques. Can. J. Rem.
Sens. 29 (3), 381387.
Qi, J., Chehbouni, A., Huete, A., Kerr, Y., Sorooshian, S., 1994. A
modified soil adjusted vegetation index. Rem. Sens. Environ. 48,
119126.
Qi, J., Marsett, R., Moran, M., Goodrich, D., Heilman, P., Kerr, Y.,
Dedieu, G., Chehnouni, A., Zhang, X., 2000. Spatial and temporal
dynamics of vegetation in the San Pedro River basin area. Agric.
For. Met. 105, 5565.
Reynolds, J., Hilbert, D., Kemp, P., 1993. Ch. scaling ecophysiology
from the plant to the ecosystem: a conceptual framework. In:
Scaling Physiological Processes: Leaf to Globe, Academy Press,
Inc., New York, pp. 127140.
Richardson, A., Wiegand, C., 1977. Distinguishing vegetation from
soil background information. Photogrammetr. Eng. Rem. Sens. 43,
15411552.
Sakai, R., Fitzjarrald, D., Moore, K., 1997. Detecting leaf area and
surface resistance during transition seasons. Agric. For. Meteorol.
84, 273284.
Sellers, P., 1985. Canopy reflectance, photosynthesis and transpira-
tion. Int. J. Rem. Sens. 6, 13351372.
-
Spanner, M., Pierce, L., Running, S., Peterson, D., 1990. The season-
ality of AVHRR data of temperate coniferous forests: relation with
leaf area index. Rem. Sens. Environ. 33, 97112.
Tucker, C., 1979. Red and photographic infrared linear combinations
for monitoring vegetation. Rem. Sens. Environ. 8, 127150.
Weiss, A., Norman, J., 1985. Partitioning solar radiation into direct
and diffuse, visible and near-infrared components. Agric. For.
Meteorol. 34, 205213.
White, J., Running, S., Nemani, R., Keane, R., Ryan, K., 1997.
Measurement and remote sensing of LAI in Rocky Mountain
Montane ecosystems. Can. J. For. Res. 27, 17141727.
Zhang, P., Anderson, B., Barlow, M., Tan, B., Myneni, R., 2004.
Climate-related vegetation characteristics derived from Moderate
Resolution Imaging Spectroradiometer (MODIS) leaf area index
and normalized difference vegetation index. J. Geog. Res. 109,
d20105.
T.B. Wilson, T.P. Meyers / Agricultural and Forest Meteorology 144 (2007) 160179 179
Determining vegetation indices from solar and photosynthetically active radiation fluxesIntroductionMethodsEstimation of spectral vegetation indicesFlux tower sitesDeciduous forestsNeedle-leaf Pine forest (BH)GrasslandsCroplands
Data analysis
ResultsPartitioning solar radiation into wavebandsIntra-annual variation of the vegetation indices at flux towersInter-annual variation of NDVIComparison of NDVILeaf area index estimated from tower-derived NDVIComparison of LAI
DiscussionsConclusionsAcknowledgmentsReferences